Prognostic significance of GAD1 overexpression in patients with resected lung adenocarcinoma

Abstract Background and Objectives In a previous genome‐wide screening, we identified hypermethylated CpG islands around glutamate decarboxylase 1 (GAD1) in lung adenocarcinoma (LADC). In this study, we aimed to investigate the methylation and expression status of GAD1 and its prognostic value in patients with LADC. Methods GAD1 methylation and mRNA expression status were analyzed using 33 tumorous and paired non‐tumorous LADC samples and publicly available datasets. The prognostic value of GAD1 overexpression was investigated using publicly available datasets of mRNA levels and 162 cases of LADC by immunohistochemistry. Results The methylation and mRNA expression levels of GAD1, each having a positive correlation, were significantly higher in LADC tumors than in paired non‐tumorous tissues. LADC patients with higher GAD1 mRNA expression showed significantly poorer prognosis for overall survival in publicly available datasets. Higher immunoreactivity of GAD1 was significantly associated with the pathological stage, pleural invasion, lymph vessel invasion, and poorer prognosis for cancer‐specific and disease‐free survival. Multivariate analysis revealed that GAD1 protein overexpression is an independent prognosticator for disease‐free survival. Conclusions GAD1 mRNA and protein expression levels were significant prognostic factors in LADC, suggesting that they might be useful biomarkers to stratify patients with worse clinical outcomes after resection.


| INTRODUCTION
Lung adenocarcinoma (LADC) is the predominant histological subtype of lung cancer and has the highest mortality rate worldwide. 1,2 Although progress in the treatment of LADC has improved short-term survival, the impacts on long-term survival remain modest. 3 Therefore, a better understanding of the mechanisms of LADC tumor progression is needed and useful prognostic molecular markers for accurately predicting the clinical outcomes of LADC are of great clinical significance.
To identify genes in the tumor that are specifically methylated at an early-stage of LADC, we had previously performed a genome-wide screening of aberrantly methylated CpG islands (CGIs) using paired tumorous and non-tumorous tissues of early-stage LADC, and identified TRIM58 as a novel candidate tumor-suppressor gene for this disease. 4 Through this screening, the glutamate decarboxylase 1 gene (GAD1) was found to be nearby hypermethylated CGIs in LADC. Because paradoxical hypermethylation-associated overexpression of GAD1 was reported recently in colorectal and liver cancers 5 and GAD1 overexpression has been reported in various neoplastic tissues, such as oral, nasopharyngeal, colorectal, liver, and gastric cancers, [5][6][7][8][9] we focused on GAD1 as a potential LADC-related gene in the present study. Moreover, the methylation and expression status and clinicopathological significance of GAD1 in LADC tumorigenesis have also not been examined previously.
Therefore, in the present study, we investigated the DNA methylation and mRNA and protein expression status of GAD1 in resected LADC tumors. Moreover, we assessed the prognostic significance of GAD1 expression in LADC using our tumor panel and publicly available datasets.

| Selection of candidate CGI
Previously obtained Human Methylation 450K array-based methylation screening data of 12 paired tumorous/nontumorous stage-I LADC sample sets from patients (6 smokers and 6 never-smokers) who underwent surgery at Tokushima University Hospital (Tokushima, Japan) between April 1999 and March 2015 were reevaluated (Table S1). 4

| Patients and tissue samples
We included tumors and non-tumorous tissues of LADC that were surgically resected at Tokushima University Hospital between April 1999 and November 2013 for additional analyses. No patients had been administered preoperative radiation, chemotherapy, or immunotherapy. For pyrosequencingbased methylation analysis and real-time PCR-based expression analysis, 33 LADC samples were used (Table S2). For immunohistochemical staining, 162 LADC samples were used (Table S3). The mean follow-up duration for the 162 patients with LADC was 48 months (range, 0.6-147 months), with 45 recurrences (27.8%) and 34 deaths (21.0%) among the patients. Tumor staging was determined based on the seventh tumornode-metastasis (TNM) classification for lung cancer. 10 The tumors were classified according to the predominant histological subtype, as proposed by the 2015 WHO classification. 11 This study was performed in accordance with the principles outlined in the Declaration of Helsinki. The ethics committee of Tokushima University Hospital approved the study (approval number 3048), and formal written consent was obtained from all patients or their representatives.

| DNA and RNA preparation and bisulfite conversion of genomic DNA
DNA and RNA were extracted using standard methods. Bisulfite conversion of DNA was conducted using the EpiTect Bisulfite Kit (QIAGEN GmbH, Hilden, Germany) following the manufacturer's instructions.

| Bisulfite pyrosequencing
Bisulfite-treated genomic DNA was amplified using a set of primers designed with PyroMark Assay Design Software version 2.0.01.15 (QIAGEN GmbH, Table S4). The target region for sequencing began 10 nucleotides (nt) before and ended 26 nt after cg15126544. PCR product pyrosequencing and methylation quantification were performed with sequencing primers using the PyroMark 24 Pyrosequencing System, version 2.0.6 (QIAGEN GmbH), according to the manufacturer's instructions.

| Real-time quantitative reversetranscription polymerase chain reaction (rqRT-PCR)
Complementary DNA was generated from isolated total RNA using the PrimeScript II first strand cDNA Synthesis Kit (TaKaRa, Shiga, Japan). rqRT-PCR was performed  Table S4) according to the manufacturers' instructions. GAPDH mRNA levels were used as internal controls for normalization. Relative expression of GAD1 mRNA was calculated using Human Lung Total RNA (TaKaRa) as a normal lung control. Research Network (http://cance rgeno me.nih.gov). mRNA expression data and DNA methylation data were available for 36 and 29 paired tumorous/non-tumorous sample sets, respectively; both types of data were available for 18 sets. Tumorous samples with mRNA expression data and survival data were available for 423 cases. Survival analyses were conducted on patients with normalized mRNA expression and overall survival (OS) profiles. Patients were divided into low-and high-expression groups according to the median GAD1 mRNA expression value.

| Data mining in bioinformatics
Kaplan-Meier Plotter (KM plotter, http://kmplot.com/ analy sis/), a publicly available online database of published microarray datasets for primary tumors with clinical information, 12 was also used to generate OS curves in 9 studies from Gene Expression Omnibus (GEO, https ://www.ncbi. nlm.nih.gov/geo/, Table S5) by setting the auto-selected best value of GAD1 mRNA expression as the cutoff. All other parameters were left at default settings.

| Immunohistochemical staining
Paraffin sections (4-µm thick) were subjected to immunohistochemical staining using the Envision system (ChemMate Envision kit; Dako, Glostrup, Denmark) according to the manufacturer's instructions. Antigen retrieval was performed by heating the dewaxed and dehydrated sections in Dako Real Target Retrieval Solution, pH 9 (Dako), using a 2100 retriever (Aptum Biologics, Ltd., Southampton, UK). A mouse anti-GAD67 monoclonal antibody (Sigma-Aldrich, St. Louis, MO, USA; G5419), diluted to 1:200 with antibody diluents (Dako), was used as the primary antibody. The proportion and intensity of GAD1 staining in the LADC samples were scored (Table S6A) independently by two different researchers.

| Statistical analysis
Student's t test or Fischer's exact test was used for comparisons between two groups. The paired t test was used for comparisons between paired samples. The relationship between continuous variables was investigated by calculating the Spearman's correlation coefficient. For survival analysis, Kaplan-Meier survival curves were constructed for groups based on univariate predictors, and differences among groups were tested with the log-rank test. Univariate and multivariate survival analyses were performed using the likelihood ratio test of the stratified Cox proportional hazard regression analysis. Differences were assessed using two-sided tests and were considered significant at a P < 0.05. Statistical analyses were performed using IBM SPSS version 24 (IBM Corporation, Armonk, NY) or the Survival package for R (https ://cran.r-proje ct.org).

CpG site within CGIs around GAD1
In a previous array-based, genome-wide methylation screening of 12 paired tumorous/non-tumorous LADC sample sets, 4 CGI-3 around GAD1 was ranked 14th as a hypermethylated CGI with a high P-value (Table S1). Because hypermethylation-associated overexpression of GAD1 was reported in colorectal and liver cancers, 5 we reevaluated the results of the array-based methylation status of each CpG site within CGI-1-4 ( Figure 1A) around GAD1 ( Figure 1B). The methylation levels of all CpG sites determined by array-based analysis within CGI-3 and in tumors were significantly higher than those in paired nontumorous tissues. Although the methylation levels in tumors were higher in CpG sites within CGI-3 than in those within CGI-4, the average β-value in non-tumor tissues was extremely and specifically low at cg15126544 and showed the largest difference of average β-value between tumor and non-tumor tissues at this site ( Figure 1B and Table S7), which is localized within the CCCTC-binding factor (CTCF)-binding site of GAD1. Similar results were observed in the Level 3 Infinium Human Methylation 450K data of 29 LADC tumors and paired non-tumor tissues from TCGA dataset ( Figure S1). Because hypermethylation around this CTCF-binding site has been reported as a possible cause of GAD1 overexpression, 5 we further assessed the methylation status of cg15126544 and GAD1 mRNA expression levels.

| Correlation between GAD1 expression and CGI methylation in LADC clinical cases
The DNA methylation status and mRNA expression status were investigated in our panel of LADC tumorous and paired non-tumorous tissues (Table S2) using pyrosequence-based methylation assays and rqRT-PCR-based expression analysis, respectively. Of the 33 sample sets, 26 (78.8%) demonstrated significantly higher methylation levels in tumor samples than in non-tumorous tissues ( Figure 1C). In the same cases, the mean GAD1 mRNA expression levels in the tumors were significantly higher than those in the paired non-tumorous tissues ( Figure 1D). There was a slightly positive (ρ = 0.251) but significant correlation between methylation levels at cg15126544 and GAD1 mRNA expression ( Figure 1E). The LADC sample set containing 18-paired samples obtained from TCGA demonstrated similar results both in methylation levels at cg15126544 and GAD1 mRNA expression ( Figure 1F,G and Figure S1). A significant and highly positive correlation between them was also observed in TCGA dataset (ρ = 0.706, Figure 1H). Because the gene expression status of cancer cells directly affects their phenotypes, including malignant features, we focused on GAD1 expression in tumors to further assess its prognostic significance in patients with LADC.

| Association of GAD1 mRNA expression levels with prognosis in LADC tumors
In our LADC cohort, a sufficient number of cases with highquality RNA suitable for expression analysis was not available for survival analysis. Therefore, to test the association between GAD1 mRNA expression levels in tumors and patients' prognosis, we first performed survival analysis of 423 patients with LADC using data obtained from TCGA dataset. The OS rate of patients with LADC with higher GAD1 mRNA expression in tumors was significantly poorer than that of patients with lower GAD1 mRNA expression in tumors ( Figure 2A). Univariate Cox regression analysis using data obtained from TCGA dataset confirmed that high GAD1 mRNA expression was associated with a worse prognostic significance for OS (Table 1). In multivariate Cox regression analysis, high GAD1 mRNA expression was also significantly associated with a poorer OS rate, suggesting that GAD1 mRNA expression is an independent prognostic factor for OS (P = 0.036, Table 1).
To validate this result, we performed survival analysis by drawing Kaplan-Meier survival curves using KM plotter ( Figure 2B). A total of 9 studies from the GEO dataset were included (Table S5). In a total of 720 patients with LADC from 9 cohorts, high GAD1 mRNA expression also significantly correlated with worse OS. In subgroup analysis of OS using datasets of KM plotter, heterogeneous results were

T A B L E 2 Correlation between GAD1
immunoreactivity and clinicopathological factors in 162 patients with LADC obtained among different cohorts. Larger cohorts such as GSE31210 and GSE50081 consistently showed that higher GAD1 mRNA expression was a poor prognostic factor, whereas cohorts with a smaller number of cases showed varying results ( Figure S2). The results of univariate Cox regression analysis confirmed these results ( Figure 2C).

| Immunohistochemical staining pattern of GAD1 and its association with prognosis in LADC tumors
To further validate the prognostic significance of GAD1 expression status, we further examined the correlation between GAD1 protein expression and clinicopathological features including prognosis in patients with LADC. We performed immunohistochemical staining of GAD1 in tissue samples from our cohort of 162 patients with LADC (Table S3). Cytoplasmic GAD1 staining was observed in LADC tumor cells with higher mRNA expression, whereas nearly no staining was observed in normal lung epithelial cells and either tumorous or non-tumorous epithelial cells in LADC with lower mRNA expression ( Figure 3A). According to the staining score (Table S6B), 112 patients (69.1%) were classified into the group with tumors showing GAD1 protein overexpression (positive GAD1 immunoreactivity). Among the various clinicopathological factors, the pathological stage, pleural invasion, and lymph vessel invasion were identified as factors significantly and positively associated with positive GAD1 immunoreactivity ( Table 2). Lymph node metastasis also tended to be more frequently observed in the positive GAD1 immunoreactivity group. According to the GAD1 protein expression status of LADC tumors, Kaplan-Meier curves of estimated OS, disease-free survival (DFS), and cancer-specific survival (CSS) were generated. Patients with GAD1 protein-overexpressing tumors showed significantly poorer DFS (P < 0.001, log-rank test) and CSS (P = 0.031, log-rank test) than those without GAD1 protein overexpressing tumors. Patients with GAD1 protein-overexpressing tumors tended to show poorer OS, although the difference between groups was not significant ( Figure 3B). Univariate Cox regression analysis confirmed that positive GAD1 immunoreactivity was significantly associated with a worse prognostic significance for DFS (Table  3). Multivariate Cox regression analysis in 162 patients revealed that GAD1 immunoreactivity was an independent prognostic factor for DFS (P = 0.011, hazard ratio = 6.424, Table 3), but not for OS and CSS (Tables S8 and S9).

| DISCUSSION
In the present study, we focused on GAD1 as a hypermethylated gene at specific CpG sites in LADC tumors and demonstrated its overexpression in tumor-specific and methylation level-associated manners in LADC. We also demonstrated the prognostic significance of GAD1 mRNA and protein expression levels in resected LADC tumors using various independent publicly available datasets and our cohort, respectively. Our study suggested that GAD1 overexpression may be a useful biomarker for predicting the prognosis of patients with LADC.
GAD1 is known to catalyze the production of γ-aminobutyric acid (GABA) from L-glutamic acid, the principal inhibitory neurotransmitter in the brain. 13,14 GAD1 overexpression has been reported in various neoplastic tissues, but not in LADC. Moreover, the associations between clinicopathological characteristics and GAD1 expression have not been well-established. The most striking finding in this study is the prognostic significance of GAD1 mRNA and protein expression in patients with LADC. Although a sufficient number of RNA samples suitable for expression analysis was not available in our cohort for survival analyses, we used various publicly available data and demonstrated that GAD1 mRNA overexpression in tumors was significantly associated with poor prognosis (OS) in independent TCGA and GEO datasets of LADC cases. In immunohistochemical analysis using our cohort, a positive cytoplasmic GAD1 staining pattern in tumor cells was significantly associated with poor prognosis, particularly DFS but not OS, in patients with LADC. Although the difference in the association between GAD1 expression and OS among datasets remains unclear, it may be explained by (a) variations in GAD1 mRNA and protein expression, (b) the smaller size of the cohort for immunohistochemical analysis compared to those of cohorts used for mRNA analysis used in our study, and (c) variations in GAD1 expression level and/or pattern among different ethnicities.
Our study also demonstrated that GAD1 protein expression in LADC was significantly associated with pleural invasion and lymph vessel invasion. These findings suggest that GAD1 overexpression might be closely associated with cellular invasion. This hypothesis is supported by previous reports of other cancers. Kimura et al 6 demonstrated that GAD1 promotes the cancer cell invasion and metastasis of oral cancer by inducing the nuclear translocation of β-catenin and secretion of MMP7, [15][16][17][18][19][20] although the regulatory mechanisms of GAD1 in β-catenin translocation remain unclear. In a brain metastasis model, it was reported that the metastatic activity of tumor cells depends on the GAD1-GABA synthesis pathway. 21 Further studies are needed to clarify the tumor-promoting activity of overexpressed GAD1.
Recently, Yan et al 5 reported hypermethylation-associated GAD1 overexpression in colorectal and liver cancers and found that this paradoxical effect was caused by the hypermethylation of the CTCF-binding site within GAD1, which may prevent CTCF binding, inhibit CTCF-mediated repressive Polycomb repressive complex 2 (PRC2) complex recruitment to the GAD1 promoter, inhibit PRC2-induced trimethylation of histone H3 lysine 27 (H3K27m3), and eliminate the blocking activity H3K27m3 for GAD1 transcription. 22,23 These observations are contradictory to the well-established paradigm that promoter DNA methylation represses transcription by inhibiting transcription factor binding and/or chromatin structure modification. [24][25][26] In this study, we also detected hypermethylation at cg15126544 within the CTCF-binding site in LADC tumors, and tumor-specific GAD1 overexpression was positively associated with hypermethylation at cg15126544 in our cohort and the TCGA dataset. Therefore, methylation of CTCF-binding sites may regulate GAD1 expression in LADC as well. However, it remains unknown whether the methylation of CGI or each CpG site around GAD1, particularly cg15126544, is the only mechanism underlying the regulation of its transcription. Interestingly, in brain metastatic tumor cells, it was reported that the downregulation of the DNA methyltransferase DNMT1 induced by the brain microenvironment-derived clusterin resulted in decreased GAD1 promoter methylation and subsequent upregulation of GAD1 expression. 21 Therefore, even the effect of methylation levels of CpG sites around GAD1 on its expression level may vary under different conditions or in different cell lineages. Indeed, MethSurv, a web tool for multivariable survival analysis using DNA methylation data obtained from TCGA datasets (https ://biit.cs.ut.ee/meths urv/), failed to show the prognostic significance of CpG sites around GAD1, including cg15126544 for OS (data not shown). Therefore, the methylation status of some CpG sites around GAD1 may contribute to its gene expression at some stages of LADC development, but not to the progression of this tumor. The GAD1 mRNA expression level data in normal lung tissues available in public databases, such as the NIH Genotype-Tissue Expression Project (https ://www.gtexp ortal.org/), as well as our immunohistochemical staining results revealed no or low GAD1 expression in normal lung tissue, suggesting that GAD1 is specifically expressed in tumor cells and contributes to the progression of tumors in LADC. Because the gene expression status appears to more directly contribute to the establishment of clinicopathological phenotypes in tumor cells, it is necessary to investigate the detailed regulatory mechanisms of GAD1 expression in LADC cells at each developmental stage of the tumor.
There are some limitations to this study. First, we demonstrated the prognostic impact of GAD1 mRNA and protein statuses mainly in Caucasian and Japanese (Asian) populations, respectively, but no data are available to directly compare GAD1 mRNA and protein expression levels among different ethnicities. Because it has been reported that the frequency of acquired alterations, such as epidermal growth factor receptor mutation, in lung tumors can vary across different ethnicities, [27][28][29] it is possible that the GAD1 expression pattern and/or levels differ between Caucasian and Asian populations. However, the prognostic significance of the GAD1 mRNA expression status in Japanese cases with LADC was demonstrated by GSE31210 in GEO datasets ( Figure 2C and Figure S2). Meta-analysis using 9 GEO datasets, including GSE31210 and 8 other studies from western countries (Table S5) also revealed the prognostic significance of the GAD1 mRNA expression status ( Figure 2C), suggesting that GAD1 overexpression is a common prognostic factor in various populations. Second, our patient cohort was relatively small even for immunohistochemical analysis, and a sufficient number of samples were not available for mRNA expression analysis to perform survival analysis. Prospective multiinstitutional studies are needed to further validate the prognostic value of GAD1 overexpression in patients with LADC.

| CONCLUSION
GAD1 overexpression appears to be a significant and independent prognostic indicator in patients with resected LADC at both the mRNA and protein levels. This information may be helpful for identifying patients at high risk of recurrence and overall survival after tumor resection of LADC.